1 // SPDX-License-Identifier: GPL-2.0-only
3 #include <linux/ceph/ceph_debug.h>
5 #include <linux/backing-dev.h>
6 #include <linux/ctype.h>
8 #include <linux/inet.h>
10 #include <linux/module.h>
11 #include <linux/mount.h>
12 #include <linux/fs_context.h>
13 #include <linux/fs_parser.h>
14 #include <linux/sched.h>
15 #include <linux/seq_file.h>
16 #include <linux/slab.h>
17 #include <linux/statfs.h>
18 #include <linux/string.h>
21 #include "mds_client.h"
24 #include <linux/ceph/ceph_features.h>
25 #include <linux/ceph/decode.h>
26 #include <linux/ceph/mon_client.h>
27 #include <linux/ceph/auth.h>
28 #include <linux/ceph/debugfs.h>
31 * Ceph superblock operations
33 * Handle the basics of mounting, unmounting.
39 static void ceph_put_super(struct super_block *s)
41 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
44 ceph_mdsc_close_sessions(fsc->mdsc);
47 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
49 struct ceph_fs_client *fsc = ceph_inode_to_client(d_inode(dentry));
50 struct ceph_mon_client *monc = &fsc->client->monc;
51 struct ceph_statfs st;
56 if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
57 data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
59 data_pool = CEPH_NOPOOL;
63 err = ceph_monc_do_statfs(monc, data_pool, &st);
68 buf->f_type = CEPH_SUPER_MAGIC; /* ?? */
71 * express utilization in terms of large blocks to avoid
72 * overflow on 32-bit machines.
74 * NOTE: for the time being, we make bsize == frsize to humor
75 * not-yet-ancient versions of glibc that are broken.
76 * Someday, we will probably want to report a real block
77 * size... whatever that may mean for a network file system!
79 buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
80 buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
83 * By default use root quota for stats; fallback to overall filesystem
84 * usage if using 'noquotadf' mount option or if the root dir doesn't
85 * have max_bytes quota set.
87 if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
88 !ceph_quota_update_statfs(fsc, buf)) {
89 buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
90 buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
91 buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
94 buf->f_files = le64_to_cpu(st.num_objects);
96 buf->f_namelen = NAME_MAX;
98 /* Must convert the fsid, for consistent values across arches */
99 mutex_lock(&monc->mutex);
100 fsid = le64_to_cpu(*(__le64 *)(&monc->monmap->fsid)) ^
101 le64_to_cpu(*((__le64 *)&monc->monmap->fsid + 1));
102 mutex_unlock(&monc->mutex);
104 buf->f_fsid.val[0] = fsid & 0xffffffff;
105 buf->f_fsid.val[1] = fsid >> 32;
110 static int ceph_sync_fs(struct super_block *sb, int wait)
112 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
115 dout("sync_fs (non-blocking)\n");
116 ceph_flush_dirty_caps(fsc->mdsc);
117 dout("sync_fs (non-blocking) done\n");
121 dout("sync_fs (blocking)\n");
122 ceph_osdc_sync(&fsc->client->osdc);
123 ceph_mdsc_sync(fsc->mdsc);
124 dout("sync_fs (blocking) done\n");
135 Opt_caps_wanted_delay_min,
136 Opt_caps_wanted_delay_max,
138 Opt_readdir_max_entries,
139 Opt_readdir_max_bytes,
146 /* string args above */
154 Opt_require_active_mds,
160 enum ceph_recover_session_mode {
161 ceph_recover_session_no,
162 ceph_recover_session_clean
165 static const struct constant_table ceph_param_recover[] = {
166 { "no", ceph_recover_session_no },
167 { "clean", ceph_recover_session_clean },
171 static const struct fs_parameter_spec ceph_mount_parameters[] = {
172 fsparam_flag_no ("acl", Opt_acl),
173 fsparam_flag_no ("asyncreaddir", Opt_asyncreaddir),
174 fsparam_s32 ("caps_max", Opt_caps_max),
175 fsparam_u32 ("caps_wanted_delay_max", Opt_caps_wanted_delay_max),
176 fsparam_u32 ("caps_wanted_delay_min", Opt_caps_wanted_delay_min),
177 fsparam_u32 ("write_congestion_kb", Opt_congestion_kb),
178 fsparam_flag_no ("copyfrom", Opt_copyfrom),
179 fsparam_flag_no ("dcache", Opt_dcache),
180 fsparam_flag_no ("dirstat", Opt_dirstat),
181 fsparam_flag_no ("fsc", Opt_fscache), // fsc|nofsc
182 fsparam_string ("fsc", Opt_fscache), // fsc=...
183 fsparam_flag_no ("ino32", Opt_ino32),
184 fsparam_string ("mds_namespace", Opt_mds_namespace),
185 fsparam_flag_no ("poolperm", Opt_poolperm),
186 fsparam_flag_no ("quotadf", Opt_quotadf),
187 fsparam_u32 ("rasize", Opt_rasize),
188 fsparam_flag_no ("rbytes", Opt_rbytes),
189 fsparam_u32 ("readdir_max_bytes", Opt_readdir_max_bytes),
190 fsparam_u32 ("readdir_max_entries", Opt_readdir_max_entries),
191 fsparam_enum ("recover_session", Opt_recover_session, ceph_param_recover),
192 fsparam_flag_no ("require_active_mds", Opt_require_active_mds),
193 fsparam_u32 ("rsize", Opt_rsize),
194 fsparam_string ("snapdirname", Opt_snapdirname),
195 fsparam_string ("source", Opt_source),
196 fsparam_u32 ("wsize", Opt_wsize),
200 struct ceph_parse_opts_ctx {
201 struct ceph_options *copts;
202 struct ceph_mount_options *opts;
206 * Parse the source parameter. Distinguish the server list from the path.
208 * The source will look like:
209 * <server_spec>[,<server_spec>...]:[<path>]
211 * <server_spec> is <ip>[:<port>]
212 * <path> is optional, but if present must begin with '/'
214 static int ceph_parse_source(struct fs_parameter *param, struct fs_context *fc)
216 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
217 struct ceph_mount_options *fsopt = pctx->opts;
218 char *dev_name = param->string, *dev_name_end;
221 dout("%s '%s'\n", __func__, dev_name);
222 if (!dev_name || !*dev_name)
223 return invalfc(fc, "Empty source");
225 dev_name_end = strchr(dev_name, '/');
227 kfree(fsopt->server_path);
230 * The server_path will include the whole chars from userland
231 * including the leading '/'.
233 fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
234 if (!fsopt->server_path)
237 dev_name_end = dev_name + strlen(dev_name);
240 dev_name_end--; /* back up to ':' separator */
241 if (dev_name_end < dev_name || *dev_name_end != ':')
242 return invalfc(fc, "No path or : separator in source");
244 dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
245 if (fsopt->server_path)
246 dout("server path '%s'\n", fsopt->server_path);
248 ret = ceph_parse_mon_ips(param->string, dev_name_end - dev_name,
249 pctx->copts, fc->log.log);
253 fc->source = param->string;
254 param->string = NULL;
258 static int ceph_parse_mount_param(struct fs_context *fc,
259 struct fs_parameter *param)
261 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
262 struct ceph_mount_options *fsopt = pctx->opts;
263 struct fs_parse_result result;
267 ret = ceph_parse_param(param, pctx->copts, fc->log.log);
268 if (ret != -ENOPARAM)
271 token = fs_parse(fc, ceph_mount_parameters, param, &result);
272 dout("%s fs_parse '%s' token %d\n", __func__, param->key, token);
277 case Opt_snapdirname:
278 kfree(fsopt->snapdir_name);
279 fsopt->snapdir_name = param->string;
280 param->string = NULL;
282 case Opt_mds_namespace:
283 kfree(fsopt->mds_namespace);
284 fsopt->mds_namespace = param->string;
285 param->string = NULL;
287 case Opt_recover_session:
288 mode = result.uint_32;
289 if (mode == ceph_recover_session_no)
290 fsopt->flags &= ~CEPH_MOUNT_OPT_CLEANRECOVER;
291 else if (mode == ceph_recover_session_clean)
292 fsopt->flags |= CEPH_MOUNT_OPT_CLEANRECOVER;
298 return invalfc(fc, "Multiple sources specified");
299 return ceph_parse_source(param, fc);
301 if (result.uint_32 < PAGE_SIZE ||
302 result.uint_32 > CEPH_MAX_WRITE_SIZE)
304 fsopt->wsize = ALIGN(result.uint_32, PAGE_SIZE);
307 if (result.uint_32 < PAGE_SIZE ||
308 result.uint_32 > CEPH_MAX_READ_SIZE)
310 fsopt->rsize = ALIGN(result.uint_32, PAGE_SIZE);
313 fsopt->rasize = ALIGN(result.uint_32, PAGE_SIZE);
315 case Opt_caps_wanted_delay_min:
316 if (result.uint_32 < 1)
318 fsopt->caps_wanted_delay_min = result.uint_32;
320 case Opt_caps_wanted_delay_max:
321 if (result.uint_32 < 1)
323 fsopt->caps_wanted_delay_max = result.uint_32;
326 if (result.int_32 < 0)
328 fsopt->caps_max = result.int_32;
330 case Opt_readdir_max_entries:
331 if (result.uint_32 < 1)
333 fsopt->max_readdir = result.uint_32;
335 case Opt_readdir_max_bytes:
336 if (result.uint_32 < PAGE_SIZE && result.uint_32 != 0)
338 fsopt->max_readdir_bytes = result.uint_32;
340 case Opt_congestion_kb:
341 if (result.uint_32 < 1024) /* at least 1M */
343 fsopt->congestion_kb = result.uint_32;
347 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
349 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
353 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
355 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
357 case Opt_asyncreaddir:
359 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
361 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
365 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
367 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
371 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
373 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
377 #ifdef CONFIG_CEPH_FSCACHE
378 kfree(fsopt->fscache_uniq);
379 fsopt->fscache_uniq = NULL;
380 if (result.negated) {
381 fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
383 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
384 fsopt->fscache_uniq = param->string;
385 param->string = NULL;
389 return invalfc(fc, "fscache support is disabled");
393 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
395 fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
397 case Opt_require_active_mds:
399 fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
401 fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
405 fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
407 fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
411 fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
413 fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
416 if (!result.negated) {
417 #ifdef CONFIG_CEPH_FS_POSIX_ACL
418 fc->sb_flags |= SB_POSIXACL;
420 return invalfc(fc, "POSIX ACL support is disabled");
423 fc->sb_flags &= ~SB_POSIXACL;
432 return invalfc(fc, "%s out of range", param->key);
435 static void destroy_mount_options(struct ceph_mount_options *args)
437 dout("destroy_mount_options %p\n", args);
441 kfree(args->snapdir_name);
442 kfree(args->mds_namespace);
443 kfree(args->server_path);
444 kfree(args->fscache_uniq);
448 static int strcmp_null(const char *s1, const char *s2)
456 return strcmp(s1, s2);
460 * path_remove_extra_slash - Remove the extra slashes in the server path
461 * @server_path: the server path and could be NULL
463 * Return NULL if the path is NULL or only consists of "/", or a string
464 * without any extra slashes including the leading slash(es) and the
465 * slash(es) at the end of the server path, such as:
466 * "//dir1////dir2///" --> "dir1/dir2"
468 static char *path_remove_extra_slash(const char *server_path)
470 const char *path = server_path;
471 const char *cur, *end;
475 /* if the server path is omitted */
479 /* remove all the leading slashes */
483 /* if the server path only consists of slashes */
489 buf = kmalloc(len + 1, GFP_KERNEL);
491 return ERR_PTR(-ENOMEM);
496 cur = strchr(path, '/');
502 /* including one '/' */
506 memcpy(p, path, len);
509 while (cur <= end && *cur == '/')
512 } while (path < end);
517 * remove the last slash if there has and just to make sure that
518 * we will get something like "dir1/dir2"
526 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
527 struct ceph_options *new_opt,
528 struct ceph_fs_client *fsc)
530 struct ceph_mount_options *fsopt1 = new_fsopt;
531 struct ceph_mount_options *fsopt2 = fsc->mount_options;
532 int ofs = offsetof(struct ceph_mount_options, snapdir_name);
536 ret = memcmp(fsopt1, fsopt2, ofs);
540 ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
543 ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
547 p1 = path_remove_extra_slash(fsopt1->server_path);
550 p2 = path_remove_extra_slash(fsopt2->server_path);
555 ret = strcmp_null(p1, p2);
561 ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
565 return ceph_compare_options(new_opt, fsc->client);
569 * ceph_show_options - Show mount options in /proc/mounts
570 * @m: seq_file to write to
571 * @root: root of that (sub)tree
573 static int ceph_show_options(struct seq_file *m, struct dentry *root)
575 struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
576 struct ceph_mount_options *fsopt = fsc->mount_options;
580 /* a comma between MNT/MS and client options */
584 ret = ceph_print_client_options(m, fsc->client, false);
588 /* retract our comma if no client options */
592 if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
593 seq_puts(m, ",dirstat");
594 if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
595 seq_puts(m, ",rbytes");
596 if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
597 seq_puts(m, ",noasyncreaddir");
598 if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
599 seq_puts(m, ",nodcache");
600 if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
601 seq_puts(m, ",ino32");
602 if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
603 seq_show_option(m, "fsc", fsopt->fscache_uniq);
605 if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
606 seq_puts(m, ",nopoolperm");
607 if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
608 seq_puts(m, ",noquotadf");
610 #ifdef CONFIG_CEPH_FS_POSIX_ACL
611 if (root->d_sb->s_flags & SB_POSIXACL)
614 seq_puts(m, ",noacl");
617 if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
618 seq_puts(m, ",copyfrom");
620 if (fsopt->mds_namespace)
621 seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
623 if (fsopt->flags & CEPH_MOUNT_OPT_CLEANRECOVER)
624 seq_show_option(m, "recover_session", "clean");
626 if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
627 seq_printf(m, ",wsize=%u", fsopt->wsize);
628 if (fsopt->rsize != CEPH_MAX_READ_SIZE)
629 seq_printf(m, ",rsize=%u", fsopt->rsize);
630 if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
631 seq_printf(m, ",rasize=%u", fsopt->rasize);
632 if (fsopt->congestion_kb != default_congestion_kb())
633 seq_printf(m, ",write_congestion_kb=%u", fsopt->congestion_kb);
635 seq_printf(m, ",caps_max=%d", fsopt->caps_max);
636 if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
637 seq_printf(m, ",caps_wanted_delay_min=%u",
638 fsopt->caps_wanted_delay_min);
639 if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
640 seq_printf(m, ",caps_wanted_delay_max=%u",
641 fsopt->caps_wanted_delay_max);
642 if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
643 seq_printf(m, ",readdir_max_entries=%u", fsopt->max_readdir);
644 if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
645 seq_printf(m, ",readdir_max_bytes=%u", fsopt->max_readdir_bytes);
646 if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
647 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
653 * handle any mon messages the standard library doesn't understand.
654 * return error if we don't either.
656 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
658 struct ceph_fs_client *fsc = client->private;
659 int type = le16_to_cpu(msg->hdr.type);
662 case CEPH_MSG_MDS_MAP:
663 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
665 case CEPH_MSG_FS_MAP_USER:
666 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
674 * create a new fs client
676 * Success or not, this function consumes @fsopt and @opt.
678 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
679 struct ceph_options *opt)
681 struct ceph_fs_client *fsc;
686 fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
692 fsc->client = ceph_create_client(opt, fsc);
693 if (IS_ERR(fsc->client)) {
694 err = PTR_ERR(fsc->client);
697 opt = NULL; /* fsc->client now owns this */
699 fsc->client->extra_mon_dispatch = extra_mon_dispatch;
700 ceph_set_opt(fsc->client, ABORT_ON_FULL);
702 if (!fsopt->mds_namespace) {
703 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
706 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
710 fsc->mount_options = fsopt;
713 fsc->mount_state = CEPH_MOUNT_MOUNTING;
715 fsc->have_copy_from2 = true;
717 atomic_long_set(&fsc->writeback_count, 0);
721 * The number of concurrent works can be high but they don't need
722 * to be processed in parallel, limit concurrency.
724 fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
727 fsc->cap_wq = alloc_workqueue("ceph-cap", 0, 1);
731 /* set up mempools */
733 page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
734 size = sizeof (struct page *) * (page_count ? page_count : 1);
735 fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
736 if (!fsc->wb_pagevec_pool)
742 destroy_workqueue(fsc->cap_wq);
744 destroy_workqueue(fsc->inode_wq);
746 ceph_destroy_client(fsc->client);
750 ceph_destroy_options(opt);
751 destroy_mount_options(fsopt);
755 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
757 flush_workqueue(fsc->inode_wq);
758 flush_workqueue(fsc->cap_wq);
761 static void destroy_fs_client(struct ceph_fs_client *fsc)
763 dout("destroy_fs_client %p\n", fsc);
765 ceph_mdsc_destroy(fsc);
766 destroy_workqueue(fsc->inode_wq);
767 destroy_workqueue(fsc->cap_wq);
769 mempool_destroy(fsc->wb_pagevec_pool);
771 destroy_mount_options(fsc->mount_options);
773 ceph_destroy_client(fsc->client);
776 dout("destroy_fs_client %p done\n", fsc);
782 struct kmem_cache *ceph_inode_cachep;
783 struct kmem_cache *ceph_cap_cachep;
784 struct kmem_cache *ceph_cap_flush_cachep;
785 struct kmem_cache *ceph_dentry_cachep;
786 struct kmem_cache *ceph_file_cachep;
787 struct kmem_cache *ceph_dir_file_cachep;
789 static void ceph_inode_init_once(void *foo)
791 struct ceph_inode_info *ci = foo;
792 inode_init_once(&ci->vfs_inode);
795 static int __init init_caches(void)
799 ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
800 sizeof(struct ceph_inode_info),
801 __alignof__(struct ceph_inode_info),
802 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
803 SLAB_ACCOUNT, ceph_inode_init_once);
804 if (!ceph_inode_cachep)
807 ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
808 if (!ceph_cap_cachep)
810 ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
811 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
812 if (!ceph_cap_flush_cachep)
815 ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
816 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
817 if (!ceph_dentry_cachep)
820 ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
821 if (!ceph_file_cachep)
824 ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
825 if (!ceph_dir_file_cachep)
828 error = ceph_fscache_register();
835 kmem_cache_destroy(ceph_dir_file_cachep);
837 kmem_cache_destroy(ceph_file_cachep);
839 kmem_cache_destroy(ceph_dentry_cachep);
841 kmem_cache_destroy(ceph_cap_flush_cachep);
843 kmem_cache_destroy(ceph_cap_cachep);
845 kmem_cache_destroy(ceph_inode_cachep);
849 static void destroy_caches(void)
852 * Make sure all delayed rcu free inodes are flushed before we
857 kmem_cache_destroy(ceph_inode_cachep);
858 kmem_cache_destroy(ceph_cap_cachep);
859 kmem_cache_destroy(ceph_cap_flush_cachep);
860 kmem_cache_destroy(ceph_dentry_cachep);
861 kmem_cache_destroy(ceph_file_cachep);
862 kmem_cache_destroy(ceph_dir_file_cachep);
864 ceph_fscache_unregister();
868 * ceph_umount_begin - initiate forced umount. Tear down down the
869 * mount, skipping steps that may hang while waiting for server(s).
871 static void ceph_umount_begin(struct super_block *sb)
873 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
875 dout("ceph_umount_begin - starting forced umount\n");
878 fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
879 ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
880 ceph_mdsc_force_umount(fsc->mdsc);
881 fsc->filp_gen++; // invalidate open files
884 static const struct super_operations ceph_super_ops = {
885 .alloc_inode = ceph_alloc_inode,
886 .free_inode = ceph_free_inode,
887 .write_inode = ceph_write_inode,
888 .drop_inode = generic_delete_inode,
889 .evict_inode = ceph_evict_inode,
890 .sync_fs = ceph_sync_fs,
891 .put_super = ceph_put_super,
892 .show_options = ceph_show_options,
893 .statfs = ceph_statfs,
894 .umount_begin = ceph_umount_begin,
898 * Bootstrap mount by opening the root directory. Note the mount
899 * @started time from caller, and time out if this takes too long.
901 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
903 unsigned long started)
905 struct ceph_mds_client *mdsc = fsc->mdsc;
906 struct ceph_mds_request *req = NULL;
911 dout("open_root_inode opening '%s'\n", path);
912 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
914 return ERR_CAST(req);
915 req->r_path1 = kstrdup(path, GFP_NOFS);
917 root = ERR_PTR(-ENOMEM);
921 req->r_ino1.ino = CEPH_INO_ROOT;
922 req->r_ino1.snap = CEPH_NOSNAP;
923 req->r_started = started;
924 req->r_timeout = fsc->client->options->mount_timeout;
925 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
927 err = ceph_mdsc_do_request(mdsc, NULL, req);
929 struct inode *inode = req->r_target_inode;
930 req->r_target_inode = NULL;
931 dout("open_root_inode success\n");
932 root = d_make_root(inode);
934 root = ERR_PTR(-ENOMEM);
937 dout("open_root_inode success, root dentry is %p\n", root);
942 ceph_mdsc_put_request(req);
947 * mount: join the ceph cluster, and open root directory.
949 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
950 struct fs_context *fc)
953 unsigned long started = jiffies; /* note the start time */
956 dout("mount start %p\n", fsc);
957 mutex_lock(&fsc->client->mount_mutex);
959 if (!fsc->sb->s_root) {
960 const char *path, *p;
961 err = __ceph_open_session(fsc->client, started);
966 if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
967 err = ceph_fscache_register_fs(fsc, fc);
972 p = path_remove_extra_slash(fsc->mount_options->server_path);
977 /* if the server path is omitted or just consists of '/' */
982 dout("mount opening path '%s'\n", path);
984 ceph_fs_debugfs_init(fsc);
986 root = open_root_dentry(fsc, path, started);
992 fsc->sb->s_root = dget(root);
994 root = dget(fsc->sb->s_root);
997 fsc->mount_state = CEPH_MOUNT_MOUNTED;
998 dout("mount success\n");
999 mutex_unlock(&fsc->client->mount_mutex);
1003 mutex_unlock(&fsc->client->mount_mutex);
1004 return ERR_PTR(err);
1007 static int ceph_set_super(struct super_block *s, struct fs_context *fc)
1009 struct ceph_fs_client *fsc = s->s_fs_info;
1012 dout("set_super %p\n", s);
1014 s->s_maxbytes = MAX_LFS_FILESIZE;
1016 s->s_xattr = ceph_xattr_handlers;
1018 fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
1020 s->s_op = &ceph_super_ops;
1021 s->s_d_op = &ceph_dentry_ops;
1022 s->s_export_op = &ceph_export_ops;
1026 s->s_time_max = U32_MAX;
1028 ret = set_anon_super_fc(s, fc);
1035 * share superblock if same fs AND options
1037 static int ceph_compare_super(struct super_block *sb, struct fs_context *fc)
1039 struct ceph_fs_client *new = fc->s_fs_info;
1040 struct ceph_mount_options *fsopt = new->mount_options;
1041 struct ceph_options *opt = new->client->options;
1042 struct ceph_fs_client *other = ceph_sb_to_client(sb);
1044 dout("ceph_compare_super %p\n", sb);
1046 if (compare_mount_options(fsopt, opt, other)) {
1047 dout("monitor(s)/mount options don't match\n");
1050 if ((opt->flags & CEPH_OPT_FSID) &&
1051 ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
1052 dout("fsid doesn't match\n");
1055 if (fc->sb_flags != (sb->s_flags & ~SB_BORN)) {
1056 dout("flags differ\n");
1063 * construct our own bdi so we can control readahead, etc.
1065 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1067 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1071 err = super_setup_bdi_name(sb, "ceph-%ld",
1072 atomic_long_inc_return(&bdi_seq));
1076 /* set ra_pages based on rasize mount option? */
1077 sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1079 /* set io_pages based on max osd read size */
1080 sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1085 static int ceph_get_tree(struct fs_context *fc)
1087 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1088 struct super_block *sb;
1089 struct ceph_fs_client *fsc;
1091 int (*compare_super)(struct super_block *, struct fs_context *) =
1095 dout("ceph_get_tree\n");
1098 return invalfc(fc, "No source");
1100 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1101 fc->sb_flags |= SB_POSIXACL;
1104 /* create client (which we may/may not use) */
1105 fsc = create_fs_client(pctx->opts, pctx->copts);
1113 err = ceph_mdsc_init(fsc);
1117 if (ceph_test_opt(fsc->client, NOSHARE))
1118 compare_super = NULL;
1120 fc->s_fs_info = fsc;
1121 sb = sget_fc(fc, compare_super, ceph_set_super);
1122 fc->s_fs_info = NULL;
1128 if (ceph_sb_to_client(sb) != fsc) {
1129 destroy_fs_client(fsc);
1130 fsc = ceph_sb_to_client(sb);
1131 dout("get_sb got existing client %p\n", fsc);
1133 dout("get_sb using new client %p\n", fsc);
1134 err = ceph_setup_bdi(sb, fsc);
1139 res = ceph_real_mount(fsc, fc);
1144 dout("root %p inode %p ino %llx.%llx\n", res,
1145 d_inode(res), ceph_vinop(d_inode(res)));
1146 fc->root = fsc->sb->s_root;
1150 if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1151 pr_info("No mds server is up or the cluster is laggy\n");
1152 err = -EHOSTUNREACH;
1155 ceph_mdsc_close_sessions(fsc->mdsc);
1156 deactivate_locked_super(sb);
1160 destroy_fs_client(fsc);
1162 dout("ceph_get_tree fail %d\n", err);
1166 static void ceph_free_fc(struct fs_context *fc)
1168 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1171 destroy_mount_options(pctx->opts);
1172 ceph_destroy_options(pctx->copts);
1177 static int ceph_reconfigure_fc(struct fs_context *fc)
1179 sync_filesystem(fc->root->d_sb);
1183 static const struct fs_context_operations ceph_context_ops = {
1184 .free = ceph_free_fc,
1185 .parse_param = ceph_parse_mount_param,
1186 .get_tree = ceph_get_tree,
1187 .reconfigure = ceph_reconfigure_fc,
1191 * Set up the filesystem mount context.
1193 static int ceph_init_fs_context(struct fs_context *fc)
1195 struct ceph_parse_opts_ctx *pctx;
1196 struct ceph_mount_options *fsopt;
1198 pctx = kzalloc(sizeof(*pctx), GFP_KERNEL);
1202 pctx->copts = ceph_alloc_options();
1206 pctx->opts = kzalloc(sizeof(*pctx->opts), GFP_KERNEL);
1211 fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
1213 fsopt->wsize = CEPH_MAX_WRITE_SIZE;
1214 fsopt->rsize = CEPH_MAX_READ_SIZE;
1215 fsopt->rasize = CEPH_RASIZE_DEFAULT;
1216 fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
1217 if (!fsopt->snapdir_name)
1220 fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
1221 fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
1222 fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
1223 fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
1224 fsopt->congestion_kb = default_congestion_kb();
1226 fc->fs_private = pctx;
1227 fc->ops = &ceph_context_ops;
1231 destroy_mount_options(pctx->opts);
1232 ceph_destroy_options(pctx->copts);
1237 static void ceph_kill_sb(struct super_block *s)
1239 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1240 dev_t dev = s->s_dev;
1242 dout("kill_sb %p\n", s);
1244 ceph_mdsc_pre_umount(fsc->mdsc);
1245 flush_fs_workqueues(fsc);
1247 generic_shutdown_super(s);
1249 fsc->client->extra_mon_dispatch = NULL;
1250 ceph_fs_debugfs_cleanup(fsc);
1252 ceph_fscache_unregister_fs(fsc);
1254 destroy_fs_client(fsc);
1255 free_anon_bdev(dev);
1258 static struct file_system_type ceph_fs_type = {
1259 .owner = THIS_MODULE,
1261 .init_fs_context = ceph_init_fs_context,
1262 .kill_sb = ceph_kill_sb,
1263 .fs_flags = FS_RENAME_DOES_D_MOVE,
1265 MODULE_ALIAS_FS("ceph");
1267 int ceph_force_reconnect(struct super_block *sb)
1269 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1272 ceph_umount_begin(sb);
1274 /* Make sure all page caches get invalidated.
1275 * see remove_session_caps_cb() */
1276 flush_workqueue(fsc->inode_wq);
1278 /* In case that we were blacklisted. This also reset
1279 * all mon/osd connections */
1280 ceph_reset_client_addr(fsc->client);
1282 ceph_osdc_clear_abort_err(&fsc->client->osdc);
1284 fsc->blacklisted = false;
1285 fsc->mount_state = CEPH_MOUNT_MOUNTED;
1288 err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1289 CEPH_STAT_CAP_INODE, true);
1294 static int __init init_ceph(void)
1296 int ret = init_caches();
1301 ret = register_filesystem(&ceph_fs_type);
1305 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1315 static void __exit exit_ceph(void)
1317 dout("exit_ceph\n");
1318 unregister_filesystem(&ceph_fs_type);
1322 module_init(init_ceph);
1323 module_exit(exit_ceph);
1325 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1326 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1327 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1328 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1329 MODULE_LICENSE("GPL");